Most people who have hired someone to build or fix something have a story about it going wrong quietly. Here is what we do so that it does not.
Every change to this site is written up, reviewed and tested before it goes anywhere. Ten automated checks run on each change proposed to the live site, every time it is updated.
They read every page for claims we have retired, and check that every script loads, that every link within the site leads to a page that exists, that the request form sends what it says it sends, and that every style is used or kept on purpose.
If a check fails, the change is marked as failing. A broken site is worse than an old one.
A test that cannot fail is worse than no test, because it tells you everything is fine.
So each one is deliberately broken before it is trusted, and the failure is recorded. Breaking a check once is not enough, because a later change can stop a break from working. So one command breaks every check again, on a copy of the site, and shows that each one still fails for the reason it should. If it does not fail when it should, it is not a check — it is decoration.
There are eight so far, each one found the hard way and each one written into our standard.
A check that reads less than it claims to cover. One that measures its own setup instead of the thing. One that counts across a group and passes while every member is wrong. A number repeated from a summary and treated as a measurement.
Every one of those has caught us. They are written down so they catch us only once.
Every piece of work comes with a plain-language account of what was wrong, what we changed, and why. Not a report for a shelf — something you can read, keep, and hand to the next person.
If you cannot understand it, that is our writing being bad, and we would like to know.
A tool on this site, a reader that reads documents aloud, could not open a single file for fifty days. One word was wrong, and the browser threw the whole script away. Our tests read the pages, and none of them asked whether every script could load, so all of them passed. We found it on the forty-fourth day, and fixed it six days later.
We fixed the fault. Then we added a check that reads every script on the site and fails if any one of them cannot load, so the next one of those is caught by a machine rather than by a person getting unlucky.
That is the actual work. Not never making mistakes. Finding them before you do.
Tell us what you have and what it needs to do. We will look at it, tell you what we find, and quote the work before anything starts.